Microsoft Researchers Rebuild Workflow with AI at Every Step

Four researchers from Microsoft's Core AI team and Fabric team have re-architected their workflow to be more AI-native, integrating artificial intelligence into every step of their process. For them, the traditional approach of using AI for narrow tasks such as summarizing a transcript no longer suffices.

The four researchers - Ben Hanrahan, Chris Gunderson, Sara Chizari, and Bruce Philips - share their experiences in rebuilding their workflow from scratch. They point to different moments when they realized that relying on one-shot habits like pasting a transcript into a prompt was no longer sufficient.

Ben Hanrahan, principal research engineer and UX researcher, notes that he made the shift when a more capable model arrived, allowing him to automate processes across his entire research process. He built an LLM wiki and agents running on a daily loop, streamlining his report writing process from two weeks to one to three days.

Chris Gunderson, UX researcher who studies security admins and analysts, made the shift when he realized that the unit of work is no longer the prompt but the system. He now pulls inputs from across content design, product design, and research for quality analysis, ensuring that prototypes are tested with a higher quality bar.

Sara Chizari, Senior UX Researcher on the same Core AI team, aimed her system at the end of the process - what happens after the report is done. She built an agent grounded in her research that people can question in their own language, making it easier for stakeholders to translate the results into their context.

The researchers agree that rebuilding their workflow from scratch has allowed them to work more efficiently and effectively, integrating AI at every step of the process. They share their systems, each shaped by the problem they care about most, but all pointing to a common realization: that AI-native ways of working require building a system rather than relying on one-shot habits.